skip to main content

Title: Revealing General Patterns of Microbiomes That Transcend Systems: Potential and Challenges of Deep Transfer Learning
ABSTRACT A growing body of research has established that the microbiome can mediate the dynamics and functional capacities of diverse biological systems. Yet, we understand little about what governs the response of these microbial communities to host or environmental changes. Most efforts to model microbiomes focus on defining the relationships between the microbiome, host, and environmental features within a specified study system and therefore fail to capture those that may be evident across multiple systems. In parallel with these developments in microbiome research, computer scientists have developed a variety of machine learning tools that can identify subtle, but informative, patterns from complex data. Here, we recommend using deep transfer learning to resolve microbiome patterns that transcend study systems. By leveraging diverse public data sets in an unsupervised way, such models can learn contextual relationships between features and build on those patterns to perform subsequent tasks (e.g., classification) within specific biological contexts.
Authors:
; ; ; ; ; ; ; ; ; ; ; ; ;
Editors:
Kinkel, Linda
Award ID(s):
2025457
Publication Date:
NSF-PAR ID:
10386599
Journal Name:
mSystems
Volume:
7
Issue:
1
ISSN:
2379-5077
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract Garnet is a common metamorphic and igneous mineral with extensive solid solution that can be stable to mantle depths ≥400 km. High-T and/or high-P garnet may contain oriented lamellae of other minerals, most commonly simple oxides (e.g., rutile, ilmenite), apatite, and, in ultrahigh-P cases, silicates including pyroxene and amphibole. Lamellae have classically been considered to be precipitation features preserving a record of former garnet chemistry richer in the lamellae nutrients (e.g., Ti4+). Such microtextural origins in precipitation systems (e.g., alloys) have long been studied via the crystallographic orientation relationships (COR) that form between a host and a separating phase, and by the shape-preferred orientation (SPO) of the lamellae. Recently, however, alternative hypotheses to precipitation have been suggested that require emplacement of lamellae in garnet by fluids, or co-growth, overgrowth, or inheritance mechanisms. These hypotheses posit that lamellae cannot be used to study former garnet chemistry. Moreover, they predict that lamellae phases, SPO, and COR should differ widely between localities, as lamellae formation will be controlled by various local rock-specific factors such as fluid presence, fluid chemistry, or mineral growth sequence. On the other hand, if lamellae characteristics are largely consistent between localities, it likely reflects control by precipitation energetics,more »rather than external factors. There have been few comparative COR studies in geologic systems, but the integrative assessment of COR, SPO, and lamellae assemblages should fingerprint lamellae growth process. To test the precipitation and alternative hypotheses, we collected large electron backscatter diffraction (EBSD) data sets for rutile, ilmenite, and apatite lamellae in garnet from the Brimfield Schist, Connecticut (≥1000 °C metamorphism; Central Maine Terrane, U.S.A.). We analyzed these data alongside published EBSD data for rutile, ilmenite, and corundum from metapegmatites metamorphosed in the eclogite facies from the Austrian Alps (Griffiths et al. 2016). The apatite data set is the first of its kind, and reveals that apatite preferentially aligns its close-packed direction parallel to that of garnet (c-axisapatite//<111>garnet). We also recognize a rutile-garnet COR related to those in meteorites with Widmanstätten patterns that are unequivocal products of exsolution. This is the first identification of direct similarities between silicate-oxide and metal-metal COR of which we are aware. Significantly, this rutile-garnet COR is found in diverse geologic settings including Connecticut and Idaho (U.S.A.), Austria, Germany, Greece, and China over a broad range of bulk-rock compositions. Results for all lamellae minerals show that COR are largely consistent between localities and, furthermore, are shared between apatite, ilmenite, and corundum. Moreover, between 70% and 95% of lamellae have COR and there is a dominant COR for each lamellae phase. Calculations show that d-spacing ratios of host-lamellae pairs can successfully predict the most commonly observed specific COR (those COR with two or more axial alignments with the host). These results, especially similarity of COR from markedly different geologic settings and a low diversity of lamellae minerals, are fully consistent with lamellae formation by precipitation (likely via exsolution). In contrast, the alternative hypotheses remain unsupported by COR results as well as by mineralogical and petrological evidence. Lamellae with similar traits as those in this work should thus be considered precipitates formed during unmixing of garnet compositions originally stable at elevated or extreme pressures and temperatures.« less
  2. Klassen, Jonathan L. (Ed.)
    ABSTRACT Omnivorous animals, including humans, harbor diverse, species-rich gut communities that impact their growth, development, and homeostasis. Model invertebrates are broadly accessible experimental platforms that enable linking specific species or species groups to host phenotypes, yet often their specialized diets and distinct gut microbiota make them less comparable to human and other mammalian and gut communities. The omnivorous cockroach Periplaneta americana harbors ∼4 × 10 2 bacterial genera within its digestive tract and is enriched with taxa commonly found in omnivorous mammals (i.e., Proteobacteria, Bacteroidetes , and Firmicutes ). These features make P. americana a valuable platform for identifying microbe-mediated host phenotypes with potential translations to mammals. Rearing P. americana insects under germfree conditions resulted in prolonging development time by ∼30% and an up to ∼8% reduction in body size along three dimensions. Germfree rearing resulted in downregulation of gene networks involved in growth, energy homeostasis, and nutrient availability. Reintroduction of a defined microbiota comprised of a subset of P. americana commensals to germfree insects did not recover normal growth and developmental phenotypes or transcriptional profiles observed in conventionally reared insects. These results are in contrast with specialist-feeding model insects (e.g., Drosophila ), where introduction of a single endemic bacterial species tomore »germfree condition-reared specimens recovered normal host phenotypes. These data suggest that understanding microbe-mediated host outcomes in animals with species-rich communities should include models that typically maintain similarly diverse microbiomes. The dramatic transcriptional, developmental, and morphological phenotypes linked to gut microbiome status in this study illustrates how microbes are key players in animal growth and evolution. IMPORTANCE Broadly accessible model organisms are essential for illustrating how microbes are engaged in the growth, development, and evolution of animals. We report that germfree rearing of omnivorous Periplaneta americana cockroaches resulted in growth defects and severely disrupted gene networks that regulate development, which highlights the importance of gut microbiota in these host processes. Absence of gut microbiota elicited a starvation-like transcriptional response in which growth and development were inhibited while nutrient scavenging was enhanced. Additionally, reintroduction of a subset of cockroach gut bacterial commensals did not broadly recover normal expression patterns, illustrating that a particular microbiome composition may be necessary for normal host development. Invertebrate microbiota model systems that enable disentangling complex, species-rich communities are essential for linking microbial taxa to specific host phenotypes.« less
  3. Abstract The use of recreational ecosystem services is highly dependent on the surrounding environmental and climate conditions. Due to this dependency, future recreational opportunities provided by nature are at risk from climate change. To understand how climate change will impact recreation we need to understand current recreational patterns, but traditional data is limited and low resolution. Fortunately, social media data presents an opportunity to overcome those data limitations and machine learning offers a tool to effectively use that big data. We use data from the social media site Flickr as a proxy for recreational visitation and random forest to model the relationships between social, environmental, and climate factors and recreation for the peak season (summer) in California. We then use the model to project how non-urban recreation will change as the climate changes. Our model shows that current patterns are exacerbated in the future under climate change, with currently popular summer recreation areas becoming more suitable and unpopular summer recreation areas becoming less suitable for recreation. Our model results have land management implications as recreation regions that see high visitation consequently experience impacts to surrounding ecosystems, ecosystem services, and infrastructure. This information can be used to include climate change impactsmore »into land management plans to more effectively provide sustainable nature recreation opportunities for current and future generations. Furthermore, our study demonstrates that crowdsourced data and machine learning offer opportunities to better integrate socio-ecological systems into climate impacts research and more holistically understand climate change impacts to human well-being.« less
  4. Abstract Animal-related outages (AROs) are a prevalent form of outages in electrical distribution systems. Animal-infrastructure interactions vary across species and regions, underlining the need to study the animal-outage relationship in more species and diverse systems. Animal activity has been an indicator of reliability in the electrical grid system by describing temporal patterns in AROs. However, these ARO models have been limited by a lack of available species activity data, instead approximating activity based on seasonal patterns and weather dependency in ARO records and characteristics of broad taxonomic groups, e.g. squirrels. We highlight available resources to fill the ecological data gap limiting joint analyses between ecology and energy sectors. Species distribution modeling (SDM), a common technique to model the distribution of a species across geographic space and time, paired with community science data, provided us with species-specific estimates of activity to analyze alongside spatio-temporal patterns of ARO severity. We use SDM estimates of activity for multiple outage-prone bird species to examine whether diverse animal activity patterns were important predictors of ARO severity by capturing existing variation within animal-outage relationships. Low dimensional representation and single patterns of bird activity were important predictors of ARO severity in Massachusetts. However, both patterns of summermore »migrants and overwintering species showed some degree of importance, indicating that multiple biological patterns could be considered in future models of grid reliability. Making the best available resources from quantitative ecology known to outside disciplines can allow for more interdisciplinary data analyses between ecological and non-ecological systems. This can result in further opportunities to examine and validate the relationships between animal activity and grid reliability in diverse systems.« less
  5. Mistletoe spatial patterns are poorly understood on a macroecological scale. Previous research conducted at the family-level on Loranthaceae from Australia demonstrated that unlike most plants, mistletoe species richness patterns do not correlate significantly with water and energy input. However, field studies suggested a relationship between the structure of the host-parasite union (haustorium) and environment. We hypothesize that haustorial type influences relationships between the abiotic environment and mistletoe spatial patterns. To investigate this hypothesis, we constructed ecological niche models for individual haustorial types. We have previously compared the distributions of haustorial types in both geographic and environmental space using geographic mapping and PCA, respectively. Here, we expand on our study by examining species richness, constructing predictive models, and emphasizing habitat types. Using the haustorial specimen collection housed at the UC Herbarium and relevant literature, we identified the haustorial type of 55 of the 73 Australia Loranthaceae mistletoe species. Using geographic distributional data from the Atlas of Living Australia and environmental data from WorldClim, we plotted haustorial groups in both geographic and environmental space, compared clusters in principle component space, and calculated Hutchinsonian niche overlap. We used regression to analyze the relationship between species richness and environmental variables at the haustorial level.more »Lastly, we constructed maximum entropy models to estimate the probability of occurrence of each haustorial group, analyzing the relative contributions of each variable to each model. We discovered that haustorial type is relatively conserved among the Australian Loranthaceae mistletoe genera, with seven out of nine genera exhibiting one haustorial type. Species with epicortical roots (ER), the ancestral character, are exclusively associated with coastal regions while those with derived haustorial types occur across the continent, including desert regions. Environmental analyses confirmed that species with ER are found in regions with milder temperatures and higher precipitation than derived types. Species richness patterns of some haustorial types, including ER, are significantly correlated with most environmental variables, while derived haustorial types are not. Maxent models for species with ER haustoria predict the highest probability of occurrence for coastal regions, while models constructed for derived types feature less bias for coastal regions. Our models demonstrate that relationships between the abiotic environment and mistletoe spatial patterns depend in part on the haustorial type. Hypotheses proposed to explain relationships between abiotic constraint on distribution and haustorial type include differences in water uptake efficiency, exposure of haustoria to the environment, longevity of haustoria, and host preference of species.« less